How an agent screens a company before a deal or a contract
An agent screens a counterparty by calling five wires in order: SEC filings to establish identity, SEC and FTC enforcement, Department of Justice actions, federal court opinions, and federal contract awards to corroborate operating scale. Five calls cost $0.045 and complete in one round trip.
/mcp. Nothing here needs an API key or a signup.
The order is the part that carries the value. Run these wires in the wrong sequence and the same five results support a different, wrong conclusion.
Identity first, because absence means two different things
Start with SEC filings. The first call is not looking for red flags, it is establishing whether the counterparty is an SEC registrant at all. That single fact changes how every later result should be read.
If the company files with the SEC, then a public record of its disclosures exists, and a search returning nothing on the enforcement and court wires is weak evidence of a clean history. If the company does not file, no such record was ever created, and the same empty result is evidence of nothing at all. An agent that skips this step and reports "no adverse findings" for a private LLC has stated a conclusion its data cannot support.
Volume is why this is a search rather than a read. EDGAR's daily index for 6 August 2026 lists 6,853 filings accepted that day, 703 of them 8-K current reports. No agent reads that. It queries it.
The sequence, and why each step follows the last
| Step | Wire | Route | What it settles | Price |
|---|---|---|---|---|
| 1 | SEC filings | /edgar/filings | Is there a public record | $0.01 |
| 2 | SEC and FTC | /enforcement/latest | Regulatory action | $0.01 |
| 3 | Justice Dept | /doj/latest | Criminal or civil action | $0.005 |
| 4 | Federal courts | /court/opinions | Litigated and decided | $0.01 |
| 5 | Federal awards | /contracts/latest | Does it win and perform work | $0.01 |
Steps 2 and 3 come before step 4 because a regulator announcing an action is faster and plainer than a court opinion, and because most enforcement resolves by settlement without ever producing a published opinion. Searching opinions first and finding nothing invites the wrong inference.
Step 4 catches what the agencies do not announce: private litigation, appeals, and cases where the company is the plaintiff rather than the defendant. That last distinction matters and is easy to lose. Being named in an opinion is not an adverse finding.
Step 5 is corroboration rather than risk. Federal award records carry the winner, the amount, the awarding agency and the place of performance, which is independent evidence that an entity does the work it claims at the scale it claims. For a subcontracting or lending decision, that is often the most useful of the five. The mechanics are in a federal contract awards API for AI agents.
The whole chain, one call at a time
# discovery only, no items: browse every wire, its filters and its price claude mcp add --transport http botwire https://thebotwire.com/mcp # 1. identity and disclosures ($0.01) curl -i "https://thebotwire.com/edgar/filings?q=Acme+Industries&since=7d&limit=25" # 2. SEC and FTC enforcement ($0.01) curl -i "https://thebotwire.com/enforcement/latest?q=Acme+Industries&since=7d" # 3. Department of Justice ($0.005) curl -i "https://thebotwire.com/doj/latest?q=Acme+Industries&since=7d" # 4. federal court opinions ($0.01) curl -i "https://thebotwire.com/court/opinions?q=Acme+Industries&since=7d" # 5. federal contract awards ($0.01) curl -i "https://thebotwire.com/contracts/latest?q=Acme+Industries&since=14d" # five calls, $0.045 total, USDC on Base, no API key and no signup
All five take q, since and limit (1 to 50). Filters differ: form on EDGAR, agency on enforcement, src on DOJ, type on courts and size on contracts. Filing search is covered in can ChatGPT or Codex read live SEC filings, and the two regulator wires are compared in how agents monitor SEC and FTC enforcement actions. Every route, filter and price is machine-readable at /llms-full.txt.
What this chain cannot tell you
Four gaps, stated plainly, because a reader who hits one unwarned does not come back.
No entity resolution. Every call above is ranked keyword matching over headline and summary text. A company that operates under a trading name, was renamed after an acquisition, or shares a common word with hundreds of unrelated entities will not resolve cleanly. Matching a legal entity across sources needs an identifier, not a string. Legal Entity Identifiers and state incorporation registries are where that work belongs.
No sanctions screening. None of these wires check watchlists. Sanctions screening runs against the Treasury OFAC lists, and it is a legal requirement rather than a research step. Do it separately, at the source, every time.
Federal courts only, and not all of them. The court wire reads the Supreme Court, the Second, Ninth and Federal Circuits. District court dockets, the other circuits and every state court are outside it. Most litigation involving most companies is in none of the four. See how agents read federal court opinions for the boundary in detail.
Short windows. The enforcement, DOJ and court wires retain 14 days. A consent decree from 2023 will not appear in any of them. These wires answer "what happened lately", not "what is this company's history". Historical screening needs the agencies' own archives, and an agent should say which period it actually searched.
FAQ
Is five calls the right depth for every screen?
No. For a low-value counterparty, steps 1 and 3 answer most of the question for $0.015. For a federal subcontracting decision, step 5 carries more weight than steps 2 to 4 combined. The sequence is a default, and an agent with a budget cap should spend it on the wires that match the decision rather than running all five out of habit.
Can an agent use this to answer "is this company legitimate"?
Not as stated. The chain surfaces public federal records within a short window. It does not verify incorporation, licensing, insurance, ownership or solvency, and it cannot prove a negative. The defensible output is a list of what was searched, what was found, and what was not covered.
Why not just run one search across everything?
Because the sources answer different questions and carry different weight. A DOJ press release and a Federal Circuit opinion are not interchangeable evidence, and collapsing them into one ranked list hides which is which. Separate calls keep the provenance attached to each result, which is what makes the output citable.
What does the whole chain cost at volume?
Five calls at $0.045 per company. Screening 1,000 counterparties costs $45 in USDC on Base with no subscription and no minimum. Per-call pricing and the reasoning behind it are in how much it costs for an AI agent to query live data.
Sources
The filing volume above comes from the SEC's own EDGAR dissemination index: the daily form index for 6 August 2026 lists 6,853 filings accepted that day, 703 of which are form 8-K. Poll cadence and per-call pricing for this service are published live at thebotwire.com/health.
Related: Can ChatGPT or Codex read live SEC filings? · How do AI agents monitor SEC and FTC enforcement actions?